diff --git a/tools/cli/cli-context.cpp b/tools/cli/cli-context.cpp
index 646c5daf3a..d432bdc7a2 100644
--- a/tools/cli/cli-context.cpp
+++ b/tools/cli/cli-context.cpp
@@ -12,10 +12,24 @@
 #include <map>
 #include <set>
 
-std::atomic<bool> g_cli_interrupted = false;
+struct cli_context_impl {
+    json messages      = json::array();
+    json pending_media = json::array(); // staged multimodal content parts
+};
+
+cli_context::cli_context(const common_params & params) : params(params), impl(new cli_context_impl()) {}
+
+cli_context::~cli_context() {
+    shutdown();
+}
+
+std::atomic<bool> & cli_context::interrupted() {
+    static std::atomic<bool> flag = false;
+    return flag;
+}
 
 static bool should_stop() {
-    return g_cli_interrupted.load();
+    return cli_context::interrupted().load();
 }
 
 static constexpr size_t FILE_GLOB_MAX_RESULTS = 100;
@@ -208,7 +222,7 @@ bool cli_context::list_and_ask_models() {
 
 void cli_context::add_system_prompt() {
     if (!params.system_prompt.empty()) {
-        messages.push_back({
+        impl->messages.push_back({
             {"role",    "system"},
             {"content", params.system_prompt}
         });
@@ -217,18 +231,18 @@ void cli_context::add_system_prompt() {
 
 void cli_context::push_user_message(const std::string & text) {
     json content;
-    if (pending_media.empty()) {
+    if (impl->pending_media.empty()) {
         content = text;
     } else {
         // multimodal message: media parts first, then the text
-        content = pending_media;
+        content = impl->pending_media;
         content.push_back({
             {"type", "text"},
             {"text", text}
         });
-        pending_media = json::array();
+        impl->pending_media = json::array();
     }
-    messages.push_back({
+    impl->messages.push_back({
         {"role",    "user"},
         {"content", content}
     });
@@ -245,7 +259,7 @@ bool cli_context::stage_media_file(const std::string & fname, const std::string
     if (type == "audio") {
         std::string ext = std::filesystem::path(fname).extension().string();
         std::transform(ext.begin(), ext.end(), ext.begin(), ::tolower);
-        pending_media.push_back({
+        impl->pending_media.push_back({
             {"type", "input_audio"},
             {"input_audio", {
                 {"data",   encoded},
@@ -253,7 +267,7 @@ bool cli_context::stage_media_file(const std::string & fname, const std::string
             }}
         });
     } else if (type == "video") {
-        pending_media.push_back({
+        impl->pending_media.push_back({
             {"type", "input_video"},
             {"input_video", {
                 {"data", encoded}
@@ -261,7 +275,7 @@ bool cli_context::stage_media_file(const std::string & fname, const std::string
         });
     } else {
         // the server detects the actual image type from the data
-        pending_media.push_back({
+        impl->pending_media.push_back({
             {"type", "image_url"},
             {"image_url", {
                 {"url", "data:image/unknown;base64," + encoded}
@@ -273,7 +287,7 @@ bool cli_context::stage_media_file(const std::string & fname, const std::string
 
 bool cli_context::generate_completion(std::string & assistant_content, cli_timings & timings) {
     json body = {
-        {"messages",          messages},
+        {"messages",          impl->messages},
         {"stream",            true},
         // in order to get timings even when we cancel mid-way
         {"timings_per_token", true},
@@ -317,7 +331,7 @@ bool cli_context::generate_completion(std::string & assistant_content, cli_timin
         }
     });
 
-    g_cli_interrupted.store(false);
+    cli_context::interrupted().store(false);
 
     if (!err.is_null()) {
         ui::show_error(format_error_message(err));
@@ -414,7 +428,7 @@ int cli_context::run() {
         }
 
         if (should_stop()) {
-            g_cli_interrupted.store(false);
+            cli_context::interrupted().store(false);
             break;
         }
 
@@ -434,19 +448,19 @@ int cli_context::run() {
         if (string_starts_with(buffer, "/exit")) {
             break;
         } else if (string_starts_with(buffer, "/regen")) {
-            if (messages.size() >= 2) {
-                size_t last_idx = messages.size() - 1;
-                messages.erase(last_idx);
+            if (impl->messages.size() >= 2) {
+                size_t last_idx = impl->messages.size() - 1;
+                impl->messages.erase(last_idx);
                 add_user_msg = false;
             } else {
                 ui::show_error("No message to regenerate.");
                 continue;
             }
         } else if (string_starts_with(buffer, "/clear")) {
-            messages.clear();
+            impl->messages.clear();
             add_system_prompt();
 
-            pending_media = json::array();
+            impl->pending_media = json::array();
             ui::show_message("Chat history cleared.");
             continue;
         } else if (
@@ -532,7 +546,7 @@ int cli_context::run() {
         cli_timings timings;
         std::string assistant_content;
         generate_completion(assistant_content, timings);
-        messages.push_back({
+        impl->messages.push_back({
             {"role",    "assistant"},
             {"content", assistant_content}
         });
diff --git a/tools/cli/cli-context.h b/tools/cli/cli-context.h
index a4d504227b..79694fb6c0 100644
--- a/tools/cli/cli-context.h
+++ b/tools/cli/cli-context.h
@@ -6,6 +6,7 @@
 #include "cli-server.h"
 
 #include <atomic>
+#include <memory>
 #include <optional>
 #include <string>
 
@@ -14,8 +15,7 @@ struct cli_timings {
     double predicted_per_second = 0.0;
 };
 
-// set by the SIGINT handler; cleared once the interrupt has been handled
-extern std::atomic<bool> g_cli_interrupted;
+struct cli_context_impl;
 
 struct cli_context {
     common_params params;
@@ -23,9 +23,6 @@ struct cli_context {
     cli_client client;                // always initialized
     std::optional<cli_server> server; // only set when no --server-base is given
 
-    json messages      = json::array();
-    json pending_media = json::array(); // staged multimodal content parts
-
     // properties of the connected server
     // will be populated by fetch_server_props()
     std::string model_name;
@@ -35,10 +32,8 @@ struct cli_context {
     bool has_audio  = false;
     bool has_video  = false;
 
-    cli_context(const common_params & params) : params(params) {}
-    ~cli_context() {
-        shutdown();
-    }
+    cli_context(const common_params & params);
+    ~cli_context();
 
     // connect to --server-base or spawn a local llama-server child;
     // argc/argv are needed to forward the server-relevant args to the child
@@ -50,6 +45,9 @@ struct cli_context {
     // stop the local server child (if any)
     void shutdown();
 
+    // set by the SIGINT handler; cleared once the interrupt has been handled
+    static std::atomic<bool> & interrupted();
+
 private:
     bool generate_completion(std::string & assistant_content, cli_timings & timings);
     void fetch_server_props();
@@ -63,4 +61,6 @@ private:
     // read a file and stage it as a multimodal content part; type is one of
     // "image", "audio", "video"; returns false if the file cannot be read
     bool stage_media_file(const std::string & fname, const std::string & type);
+
+    std::unique_ptr<cli_context_impl> impl;
 };
diff --git a/tools/cli/cli.cpp b/tools/cli/cli.cpp
index b361553ade..dcdb6aeac2 100644
--- a/tools/cli/cli.cpp
+++ b/tools/cli/cli.cpp
@@ -16,14 +16,14 @@
 
 #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
 static void signal_handler(int) {
-    if (g_cli_interrupted.load()) {
+    if (cli_context::interrupted().load()) {
         // second Ctrl+C - exit immediately
         // make sure to clear colors before exiting (not using LOG or console.cpp here to avoid deadlock)
         fprintf(stdout, "\033[0m\n");
         fflush(stdout);
         std::exit(130);
     }
-    g_cli_interrupted.store(true);
+    cli_context::interrupted().store(true);
 }
 #endif